The Archive

Empirical study on fine-tuning and metrics for neural decompilation of Dart AOT binaries with 154-task HumanEval-Dart benchmark.

Study on method-level energy consumption prediction in Java bytecode using execution time, examining limits of static metrics.

Distributed game-theoretic RL for attacker detection in integrated sensing and communication (ISAC) 6G networks.

Training-free acceleration for diffusion/flow models via x-prediction endpoint decodability without retraining or distillation.

LLM-guided measurement credibility correction for industrial soft sensing without explicit fault labels or process equations.

Stratechery proposes talking points for Zuckerberg's earnings call; commentary on Meta's AI and metaverse strategy.

Forterra has deployed more than 100

Berkeley BAIR analyzes commodity AI inference costs dropping 50-900x annually, arguing sufficient intelligence now enables agent-centric data systems.

Google expands Gemini API Managed Agents with background task execution and remote MCP support for production deployments.

sqlite-utils 4.0rc4 release candidate incorporates Claude feedback before stable launch.

Commentary on a recent model launch framed as historically significant, lacks specifics on model capabilities, architecture, or performance.

Cohere releases open-source Arabic speech recognition model for enterprise transcription across Arabic dialect variants.

Australian Payments Plus deploys ChatGPT Enterprise and Codex to accelerate payments processing with human oversight.

Meta releases Muse Image and Muse Video, generative models for instruction-following image creation, editing, and video synthesis with native audio.

Tencent releases Hy3, a 295B-param MoE model with 21B active params under Apache 2.0, claiming performance parity with 2-5x larger open-source competitors.

An AI agent carried out the technical execution of a real-world ransomware attack for the first known time, but new details show a human still chose the victim, set up the infrastructure, and supplied stolen credentials — meaning it wasn't quite the fully autonomous cybercrime…

SK Hynix is experiencing a boom credited to AI. It will ride that to a multi-billion dollar US IPO, expected to take place on Friday.

Training LLMs at massive scale brings unique infrastructure challenges, especially as jobs span thousands of GPUs and run for extended periods. The longer these... Training LLMs at massive scale brings unique infrastructure challenges, especially as jobs span thousands of GPUs…

"The reality is, when you're optimizing for production, you start looking at a price/performance," Guillermo Rauch tells TechCrunch.

The new controls let you control how quickly Siri speaks and expressive its voice is.

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. OpenAI CEO Sam Altman’s oft-discussed promise that Americans will share in the wealth AI creates was in the news again last week. On…

CamVLA: Vision-Language-Action model that learns camera calibration in-situ for robust robot manipulation under viewpoint shifts.

Label-free deep learning for astronomical transient classification using synthetic injection and contaminated data.

Weak-to-strong distillation for post-training: run RL on smaller models, distill gains to scale larger ones cheaply.

LLM-as-a-Verifier: verification as a scaling axis; fine-grained feedback for agent tasks without additional training.

SearchGen: benchmark and corpus for evaluating visual generation on out-of-distribution, long-tailed prompts via web search grounding.

Theoretical analysis of discrete diffusion model training: Oracle Distance theorem relating ELBO to reverse-process KL.

TabPack: MLP ensemble for tabular deep learning with diverse hyperparameters, reducing tuning overhead.